Ammonia is one of the key environmental parameters affecting the healthy growth of pigs. And it is the key to ensure the healthy growth of pigs by timely and accurately grasping the trend of ammonia concentration in piggeries. In order to improve the accuracy and efficiency of ammonia concentration prediction in piggeries, a prediction model of ammonia concentration in piggeries based on empirical mode decomposition and long shortterm memory neural network (EMD-LSTM) was proposed. Firstly, the sequence data of ammonia concentration was decomposed to obtain the intrinsic mode function (IMF) at different time scales. Then, the longterm memory neural network prediction model was established for the intrinsic mode function. Finally, the prediction results of the components were summed as the final value of the concentration. The prediction model proposed was applied to the prediction of ammonia concentration in a pig farm in Yixing, Jiangsu Province. In order to verify the performance of the prediction model, the prediction model was compared with Elman prediction model, recurrent neural network (RNN) prediction model, longterm memory neural network prediction model and empirical mode decomposition and recurrent neural network prediction model. The results showed that the prediction accuracy of the empirical mode decomposition and longterm memory neural network model was higher. Compared with the real values, the mean absolute error, mean absolute percentage error and root mean square error were 0.0723mg/m3,0.6257% and 0.0945mg/m, respectively.